Carbon project boundary encroachment and land-tenure conflict monitoring
Agricultural encroachment, road cuts and new settlements can quietly erode a registered carbon project area for years before an audit catches it. Dense satellite time-series, combined with boundary-aware change detection, gives registry auditors and investors a continuous record rather than a snapshot.
Sensors
- Planet Basemaps (PlanetScope): 3 m native resolution, near-daily revisit over most tropical latitudes. Dense monthly or quarterly basemaps suppress cloud artefacts through compositing, giving a clean before/after surface for perimeter differencing. Useful for detecting clearings of a few hundred square metres.
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at equatorial latitudes with both satellites. NDVI and NBR time-series from the freely available archive (2015-present) provide the statistical baseline against which new disturbances are flagged. Minimum reliably detectable clearing is roughly 0.1 ha at this resolution.
- Sentinel-1 SAR (C-band): 10 m ground range resolution in IW mode, 6-12 day revisit depending on latitude. C-band backscatter drops sharply when forest canopy is removed, and SAR penetrates cloud cover entirely. Critical for projects in persistently overcast regions where optical sensors may return no usable observations for weeks.
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral. Tasked on demand for sub-metre confirmation of flagged events: distinguishing a new agricultural plot from a windthrow gap, or identifying a structure footprint versus a natural clearing. Commercially licensed; adds cost and latency of days to weeks depending on tasking queue and cloud.
Why perimeter surveillance is not the same as deforestation alerting
Most published deforestation alert systems, including GLAD alerts from the University of Maryland and Brazil's DETER, are designed to detect large-scale forest loss across entire landscapes. They were not built to answer the specific legal question that matters to a carbon registry auditor: has activity crossed this registered polygon boundary, and if so, by how much and when?
The distinction is consequential. A clearing 500 m outside a project boundary is irrelevant to that project's accounting. A clearing of 0.3 ha that crosses the boundary line by 80 m is a disclosure event. Generic alert products do not carry that geometric precision. Boundary-aware surveillance requires registering the project polygon as a spatial constraint and running change detection relative to it, not across an undifferentiated landscape tile.
What a fresh track gives away before the clearing does
Agricultural encroachment rarely begins with a large clearing. The typical sequence is: access track cut first, then progressive clearing inward from the track. SAR backscatter is sensitive to this early-stage signal. A new unpaved track through closed-canopy forest produces a linear low-backscatter feature in C-band imagery at widths as narrow as 5-10 m, detectable in Sentinel-1 IW mode before any substantial canopy loss is visible in optical data.
Monitoring for linear features, not just area loss, therefore gives several weeks of additional lead time. This matters because the earlier an encroachment is flagged, the more credible the remediation options are. A track that has been open for two months and has not yet been farmed is a different legal situation from one that has supported three planting seasons.
Road extraction from SAR time-series is a published method class. It is not infallible: tracks under closed canopy with high soil moisture can be ambiguous, and very narrow paths below roughly 3-5 m width will fall below Sentinel-1's effective detection limit. Confirmation with sub-metre optical tasking resolves most ambiguous cases.
Dense time-series versus the audit snapshot problem
Carbon project audits are typically conducted annually or less frequently. A single-date comparison between audit visits can miss the full history of what happened inside the boundary: a clearing that was farmed and then abandoned, or a structure that was built, used and demolished. Dense time-series imagery preserves that history.
Sentinel-2's archive runs from 2015 and Planet's commercial archive extends back to roughly 2016 for many regions. For projects registered after those dates, it is possible to reconstruct a near-complete record of boundary-zone activity at monthly or better temporal resolution. This is directly useful for institutional investors conducting portfolio due diligence on projects acquired after registration: the satellite record does not depend on what the original project developer chose to disclose.
The practical limit is cloud. In humid tropical regions, monthly compositing with Sentinel-2 can still leave data gaps in the wettest months. Fusing optical composites with SAR observations, which carry no cloud penalty, substantially reduces those gaps but introduces its own complexity: SAR and optical disturbance signals are not identical, and a fusion pipeline needs to be calibrated against known-disturbance reference sites to avoid false positives.
Land-tenure conflict as a separate signal
Encroachment is not always agricultural. In regions where carbon project boundaries overlap with disputed customary land claims, settlement expansion can appear as clusters of new structure footprints rather than cleared agricultural plots. These are detectable at sub-metre resolution with WorldView-3 but are below the reliable detection threshold of Sentinel-2 at 10 m unless the settlement is large enough to produce measurable canopy loss.
Tenure conflict also manifests as repeated small-scale clearing events that individually fall below alert thresholds but cumulatively represent significant area loss. A change-detection system that flags only events above a fixed area threshold will miss this pattern. Cumulative area accounting over a rolling 12-month window, summed across all flagged events within the boundary buffer zone, is the appropriate metric for this signal.
It is worth stating plainly that satellite evidence can document the spatial and temporal pattern of encroachment but cannot determine legal ownership or the legitimacy of competing claims. That interpretation requires ground-truth and legal expertise. The satellite layer provides the factual record; the legal conclusion is drawn elsewhere.
What the method cannot do
Below roughly 0.1 ha, clearing detection with free optical data becomes unreliable. Selective removal of individual trees, understorey clearing without canopy loss, and very narrow track cuts are all below the effective detection floor of Sentinel-2. Planet's 3 m basemaps push the floor lower, to roughly 0.02-0.05 ha for well-contrasted clearings, but this requires a commercial licence and the floor is not fixed: it depends on canopy closure, soil contrast and compositing quality.
Persistent cloud in equatorial regions can delay optical detection by four to eight weeks even with dense compositing. SAR fills much of this gap but is not a complete substitute: SAR is less sensitive to low-intensity degradation such as understorey burning or selective harvest, which leave the upper canopy largely intact.
Finally, a boundary-surveillance system produces evidence of change, not attribution. Determining whether an encroachment was carried out by a specific actor, whether it constitutes a breach of the project's validation and verification standard, and what the carbon accounting consequences are, all require human judgement applied to the satellite record.
From alert to audit package
The practical output for a registry auditor or institutional investor is not a raw change-detection layer. It is a structured evidence package: georeferenced before/after image pairs, a change polygon with area and date-of-first-detection, a time-series plot showing the disturbance trajectory, and a confidence classification based on the number of independent sensor observations that corroborate the event.
Satellize structures this kind of output for clients conducting ongoing portfolio surveillance, drawing on open Sentinel and Landsat archives combined with commercial tasking where sub-metre confirmation is needed.
For registry bodies considering systematic portfolio screening, the most useful starting point is a boundary-buffer audit across the existing project portfolio: a retrospective pass over the archive to establish which projects have documented boundary-zone activity and which do not. That baseline then becomes the reference against which ongoing monitoring is measured.
Typical figures
| Optical spatial resolution | 10 m (Sentinel-2), 3 m (PlanetScope basemaps), 1.24 m multispectral / 0.31 m pan (WorldView-3) |
| SAR spatial resolution | 10 m ground range (Sentinel-1 IW mode) |
| Revisit cadence | 5 days optical (Sentinel-2, dual satellite); near-daily (PlanetScope); 6-12 days SAR (Sentinel-1, latitude-dependent) |
| Minimum reliably detectable clearing | ~0.1 ha (Sentinel-2); ~0.02-0.05 ha (PlanetScope basemaps, well-contrasted conditions); sub-0.01 ha possible with WorldView-3 confirmation |
| Cloud penetration | Full (Sentinel-1 SAR); none (optical sensors); compositing reduces but does not eliminate optical gaps in humid tropics |
| Archive depth | Sentinel-2 from 2015; Landsat from 1972 (30 m); PlanetScope from ~2016 for many regions (commercial) |
| Alert latency | 2-5 days from acquisition to processed alert for Sentinel-1/2 open data; days to weeks for WorldView-3 tasking depending on queue |
| Spectral bands used | Visible, NIR, SWIR (Sentinel-2 bands 2-8A, 11, 12); C-band VV/VH (Sentinel-1); 8-band VNIR+SWIR (WorldView-3) |
| Delivery formats | GeoTIFF change polygons, GeoJSON boundary-alert feeds, PDF evidence packages with image chips and time-series plots |
| Coverage | Global; Sentinel-1 SAR coverage varies by acquisition mode and region; check ESA coverage maps for specific project locations |
Analytics Satellize can run
| Boundary-buffer change-detection layer | Bi-temporal and multi-temporal NDVI/NBR differencing within a configurable buffer zone around the registered project polygon; statistical threshold based on pre-disturbance baseline variance | GeoTIFF and GeoJSON polygon layer of flagged change events, with area, centroid coordinates and date-of-first-detection |
| SAR-optical fusion disturbance alert | Sentinel-1 backscatter time-series combined with Sentinel-2 composites; events flagged when both sensors show consistent disturbance signal, reducing false-positive rate from cloud artefacts | Fortnightly alert feed with confidence score (single-sensor vs. corroborated) and image chip pairs |
| Linear feature (track) extraction | Directional filter and morphological analysis on SAR backscatter difference images; new linear low-backscatter features classified as candidate access tracks | GeoJSON line layer of candidate new tracks within boundary buffer, with date range of first appearance |
| Retrospective boundary-zone audit | Dense time-series reconstruction from Sentinel-2 and Landsat archive; monthly composite stack from project registration date to present; cumulative disturbance area accounting per quarter | PDF audit report with timeline of boundary-zone activity, cumulative area chart and georeferenced image evidence |
| Sub-metre confirmation imagery | Tasked WorldView-3 acquisition over flagged events; manual and automated classification of clearing type (agricultural plot, structure footprint, natural gap) | Annotated high-resolution image chip with classification label and area measurement, suitable for inclusion in registry submission |
| Portfolio-level encroachment risk ranking | Boundary-zone change rate calculated for each project in a portfolio; ranked by cumulative flagged area as percentage of registered project area over a defined look-back period | Tabular risk ranking with per-project metrics; GIS layer coloured by risk tier for portfolio map |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.